download embedding model on build time

Dockerfile

@@ -14,6 +14,9 @@ WORKDIR $HOME/src/app
 COPY --chown=user requirements.txt ./
 RUN pip install -r requirements.txt
 
+RUN huggingface-cli download sentence-transformers/all-mpnet-base-v2 \
+    --local-dir /model/all-mpnet-base-v2 --local-dir-use-symlinks False
+
 # Copy the rest of your application's code
 COPY --chown=user ./src .
 

src/conversation.py

@@ -8,6 +8,7 @@ from langchain_community.embeddings import HuggingFaceEmbeddings
 import os
 
 openai_api_key = os.environ.get("OPENAI_API_KEY")
+model_name = os.environ.get('MODEL_NAME', 'all-MiniLM-L6-v2')
 
 class Conversation_RAG:
     def __init__(self, model_name="gpt-3.5-turbo"):
@@ -15,9 +16,7 @@ class Conversation_RAG:
 
     def get_vectordb(self):
         index = pinecone.Index(os.environ.get("PINECONE_INDEX"))
-        embeddings = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-mpnet-base-v2",
-        )
+        embeddings = HuggingFaceEmbeddings(model_name=f"model/{model_name}")
         vectordb = Pinecone(index, embeddings, "text")
         
         return vectordb

src/vector_index.py

@@ -4,6 +4,8 @@ from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_community.embeddings import HuggingFaceEmbeddings
 import os, uuid
 
+model_name = os.environ.get('MODEL_NAME', 'all-MiniLM-L6-v2')
+
 def create_vector_store_index(file_path):
 
     file_path_split = file_path.split(".")
@@ -29,9 +31,7 @@ def create_vector_store_index(file_path):
 
     index = pc.Index(os.environ.get("PINECONE_INDEX"))
 
-    embeddings = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-mpnet-base-v2",
-        )
+    embeddings = HuggingFaceEmbeddings(model_name=f"model/{model_name}")
 
     batch_size = 32